The Harvard Clean Energy Project: Large-Scale ... · The Harvard Clean Energy Project: ... screening eﬀort toward an active engineering of novel organic ... ideas from machine learning,93,94

Published: August 22, 2011

r 2011 American Chemical Society 2241 dx.doi.org/10.1021/jz200866s | J. Phys. Chem. Lett. 2011, 2, 2241–2251

PERSPECTIVE

pubs.acs.org/JPCL

The Harvard Clean Energy Project: Large-Scale ComputationalScreening and Design of Organic Photovoltaics on the WorldCommunity GridJohannes Hachmann,*,† Roberto Olivares-Amaya,† Sule Atahan-Evrenk,† Carlos Amador-Bedolla,†,‡

Roel S. S�anchez-Carrera,||,† Aryeh Gold-Parker,† Leslie Vogt,† AnnaM. Brockway,§ and Al�an Aspuru-Guzik†,*†Department of Chemistry and Chemical Biology, Harvard University, 12 Oxford Street, Cambridge, Massachusetts 02138,United States‡Facultad de Química, Universidad Nacional Aut�onoma de M�exico, M�exico, DF 04510, M�exico§Department of Chemistry, Haverford College, 370 Lancaster Avenue, Haverford, Pennsylvania 19041, United States

The sun is an abundant source of energy, and its input on earthexceeds the global consumption by 4 orders of magnitude. It

is hence an obvious alternative to fossil or nuclear energy suppliesand will play an important role in safely and sustainably coveringthe rising demands of the future.1�4 The current cost of elec-tricity from commercial silicon-based solar cells is unfortunatelystill around 10 times higher than that of utility-scale electricalpower generation.5,6 These traditional inorganic photovoltaicscome with further shortcomings, such as a complicated andenergy-intensive manufacturing process that leads to high pro-duction costs. They can also contain rare or hazardous elements,and the devices tend to be heavy, bulky, rigid, and fragile.

Carbon-based solar cells have emerged as one of the interest-ing alternatives to this conventional technology.7�9 Organicphotovoltaics (OPVs) range from crystalline small moleculeapproaches10�12 and certain dye-sensitized Gr€atzel cells13 toamorphous polymers (plastics).14,15 OPVs have great potentialin two important areas: they promise simple, low-cost, and high-volume production16 as well as the prospect of merging theunique flexibility and versatility of plastics with electronic fea-tures. OPVs can be processed via roll-to-roll printing,17,18 andthere is active research in sprayable and paintable materials.19�22

Additionally, OPVs can be semitransparent,23 variously colored,24

lightweight,14 and they can essentially be molded into anyshape.25 These properties make OPVs a promising candidateto achieve the ubiquitous harvesting of solar energy,26,27 with

building-integrated28�30 and ultraportable applications31�33 asprimary targets. The Equinox Summit international committee,for example, recently suggested the use of OPVs for the basicelectrification of 2.5 billion people in rural areas without access tothe power grid.34

There are, however, still significant issues to overcome inorder to make plastic solar cells a viable technology for the future.The cardinal problems are their relatively low efficiency andlimited lifetime:35,36 the power conversion record has onlyreached 9.2%,37 and current materials still degrade when exposedto the environment.38�41 An increase in efficiency to about 10�15% in combination with lifetimes of over 10 years (for produc-tion materials) could push the power generation costs of OPVsbelow that of other currently available energy sources.42,43

In 2008, we started the Harvard Clean Energy Project(CEP)44 to help find such high-efficiency OPV materials. Thisperspective article gives a general overview of CEP and providesthe context for a series of detailed technical and result-orientedpapers to follow. In section I we introduce the motivation andoverall setup of the project, followed by a presentation of itsdifferent components: section II discusses our OPV candidateslibrary, section III explains the use of cheminformatics descriptors

Received: June 27, 2011Accepted: August 11, 2011

ABSTRACT: This Perspective introduces the Harvard Clean Energy Project (CEP), atheory-driven search for the next generation of organic solar cell materials. We give a broadoverview of its setup and infrastructure, present first results, and outline upcomingdevelopments. CEP has established an automated, high-throughput, in silico framework tostudy potential candidate structures for organic photovoltaics. The current project phase isconcerned with the characterization of millions of molecular motifs using first-principlesquantum chemistry. The scale of this study requires a correspondingly large computationalresource, which is provided by distributed volunteer computing on IBM’sWorld CommunityGrid. The results are compiled and analyzed in a reference database andwill bemade availablefor public use. In addition to finding specific candidates with certain properties, it is the goal ofCEP to illuminate and understand the structure�property relations in the domain of organicelectronics. Such insights can open the door to a rational and systematic design of future high-performance materials. The computational work in CEP is tightly embedded in a collabora-tion with experimentalists, who provide valuable input and feedback to the project.

2242 dx.doi.org/10.1021/jz200866s |J. Phys. Chem. Lett. 2011, 2, 2241–2251

The Journal of Physical Chemistry Letters PERSPECTIVE

to rapidly assess the potential of these candidates, and sectionIV is focused on the high-level calculation hierarchy in CEP.Section V is concerned with the calibration of the obtainedresults. The CEP database is introduced in section VI and theWorld Community Grid (WCG)—our primary computationalresource—in section VII. Throughout sections I�VII we alsoindicate the next stages and extensions to the current setup. Wesummarize our discussion in the last two paragraphs of the paper.

I. The Harvard Clean Energy Project. The key parameters forthe improvement of OPVs are essentially known; however,engineering materials that combine all these features is a hardproblem.45�49 Traditional experimental development is largelybased on empirical intuition or experience within a certain familyof systems, and only a few examples can be studied per year dueto long turnaround times of synthesis and characterization.50

Theoretical work is usually also restricted to a small set of candi-dates for which different aspects of the photovoltaic process aremodeled.51�53

The Clean Energy Project stands out from other computa-tional materials science approaches as it combines conventionalmodeling with strategies frommodern drug discovery:54�61 CEPfeatures an automated, high-throughput infrastructure for a syste-matic screening of millions of OPV candidates at a first-principleselectronic structure level.62 It also adopts techniques fromcheminformatics63,64 and heavily relies on data mining.65,66 Pio-neering work on cheminformatics-type approaches and massiveelectronic structure calculations was, e.g., performed by Rajanet al.67�69 and Ceder et al.,70 respectively, in the context ofinorganic solids. An in silico study combining the scale and levelof theory found in CEP is, however, unprecedented.

As the starting point for CEP, we have chosen to investigatethemolecular motifs at the heart of OPVmaterials.45,71 A suitablemotif is a necessary condition for a successful OPV development.Only a limited number of structural patterns have been exploredso far, while the endless possibilities may well hold the key toovercoming the current material issues. We emphasize that apromising molecular structure is not a sufficient condition,as condensed matter and device considerations have to beaddressed as well. CEP is set up with a calculation hierarchyin which we will successively characterize relevant electronicstructure aspects of our OPV candidates. Eventually, we will go

beyond single molecule, gas-phase studies and consider inter-molecular and bulk phase problems.

In addition to the search for specific structures with a desiredset of properties, we also try to arrive at a systematic under-standing of structure�property relationships.72�75 Learningabout underlying design principles is the key to moving from ascreening effort toward an active engineering of novel organicelectronics.76,77

While its centerpiece is the in silico study of OPV candidates,we point out that an overarching theme of CEP is also the tightintegration of experimental and theoretical work. The project isin part guided by input from experimentalist collaborators (inparticular, the Bao Group at Stanford University), and our mostpromising candidates are subject to in-depth studies in theirlaboratories.78 CEP is designed as a community tool and weinvite and welcome joint ventures.

Figure 1 summarizes the overall structure and workflow ofCEP, and in the following sections we discuss its variouscomponents.

II. Molecular Candidate Libraries. The molecular structure ofessentially all organic electronic materials features a conjugatedπ-backbone.79 Modern semiconducting compounds are oftencomposed of linked or fused (hetero)aromatic scaffolds.80

We have developed a combinatorial molecule generator tobuild the primary CEP library. It contains ∼10 000 000molecular motifs of potential interest (∼3 600 000 distinctconnectivities, each with a set of conformers), which cover smallmolecule OPVs and oligomer sequences for polymeric materi-als. They are based on 26 building blocks and bonding rules (seeFigure 2), which were chosen following advice from the BaoGroup considering promise and feasibility. The fragmentswere linked and fused up to a length of 4�5 units accordingto the given rules. The generator is graph-based and employs aSMILES (simplified molecular input line entry specification)81

string representation of the molecules as well as SMARTS(SMILES arbitrary target specification) in its engine. It is builtaround Marvin Reactor,82 and the Corina code83 providesforce-field optimized three-dimensional structures. A detaileddescription of the library generator is in preparation.

The construction of the primary library was tailored towardOPV donor candidates, but the obtained structures are of interestfor organic electronics in general.84,85 Our generator approach isflexible and can readily produce additional libraries (e.g., geared

The Clean Energy Project stands outfrom other computational materi-als science approaches as it com-bines conventional modeling withstrategies from modern drug dis-

covery: CEP features an automated,high-throughput infrastructure for asystematic screening of millions ofOPV candidates at a first-principles

electronic structure level.

Figure 1. Structure and workflow of CEP.



toward acceptor materials or Gr€atzel cell dyes) by using adifferent set of fragments and connection rules. Substituentscan be incorporated in a similar fashion. Combinatorial librariesallow for an exhaustive and systematic exploration of well-defined chemical spaces, but the number of generated moleculesgrows exponentially. At a later stage we plan to use a geneticalgorithm86�90 as a complement to the current approach.62 Itsfitness function will be based on the information gathered fromthe screening of the primary library. Finally, CEP also providesthe facility (e.g., to collaborators) to manually add structures tothe screening pool.

III. Cheminformatics Descriptors. While the main objective ofCEP is the first-principles characterization of OPV candidates,we also explore the use of cheminformatics descriptors91,92 andideas from machine learning,93,94 pattern recognition,95,96 anddrug discovery54,56,64 to rapidly gauge their quality. We havedevised descriptor models for parameters such as the short-circuitcurrent density (Jsc), the open-circuit voltage (Voc), and thepower conversion efficiency (PCE). Our models are the first stepin a successive ladder of approximations for these key quantitiesassociated with photovoltaic performance.

The basic strategy behind this approach is to identify andexploit correlations between certain physicochemical or topolo-gical descriptors and the properties of interest. Suitable descrip-tors are combined into models which are then empirically para-metrized using a training set of experimentally well-characterizedreference systems. As descriptors are easily computed, we can

quickly (i.e., within a few days on a single workstation) obtain an initialassessment and preliminary ranking of the entire molecular library.

In Figure 3 we present the linear regression model for Vocalong with the histogram of the calculated values for themolecular library. We note that our model for Voc shows a verygood correlation. This can be rationalized considering theprincipal dependence of Voc on intramolecular properties, whichare apparently well reflected in the employed molecular descrip-tors. The Jsc model behaves similarly well although Jsc is alsolinked to bulk effects. For the fill factor—a quantity primarilydetermined by morphology and device characteristics—wecould, in contrast, only obtain poor models.

The early stages of this work97 utilized descriptors from theMarvin code by ChemAxon,82 and for the modeling we em-ployed the R statistics package.98 Recently, we started using themore comprehensive descriptor set from Dragon99 and thespecialized modeling code StarDrop.100 A focus of our currentwork is to utilize the quantum chemical results discussed in thefollowing section as a descriptor basis in our models.

As in biomedical applications of cheminformatics, we do notexpect quantitative results. Rather, this technique can yield valuabletrends, which we use to prioritize and prune the high-levelscreening and to uncover molecular motifs of particular interest.In ref 97 we give an introduction to this approach with a detaileddiscussion of the systematic construction and optimization ofdescriptor models.

Figure 2. The 26 building blocks used for generating the CEP mole-cular library. TheMg atoms represent chemical handles, i.e., the reactivesites in the generation process. We introduce simple links between twomoieties (by means of substituting two Mg for a single C�C bond) aswell as the fusion of two rings.

Figure 3. Top: Linear regression model for Voc with the training setdata. Bottom: Histogram of the projected Voc values in the molecularlibrary with the position of the training set marked in green (witharbitrary height). See ref 97 for details.

A suitable molecular motif is anecessary condition for a successful

OPV development.



IV. First-Principles Screening Hierarchy. Electronic structuretheory offers a way to probe the properties of OPV materialsand the photophysical processes in organic solar cells.101,102 Thecomplexity of these problems, however, poses severe methodo-logical challenges. Multiscale simulations have improved con-siderably in recent years,51,103 but in practice we still commonlychoose to model, approximate, or deduce the different aspects ofthe problem separately. CEP adopts such a divide-and-conquerapproach and combines it with a calculation hierarchy to screenfor promising material candidates. This multilevel setup isdesigned to successively address the relevant issues in OPVsand provide results at an increasing level of theory. At each stage,the candidates are rated with respect to the investigated para-meters. The scoring is freely customizable to reflect differentresearch priorities. The most promising candidates and relatedstructures from the library receive priority in the CEP hierarchy,and their characterization is expedited. On the other hand, if acandidate is unfit with respect to a tested criterium and henceunlikely to be successful overall, it has lower priority and may becharacterized in less detail.

The early CEP stages concentrate on variousmolecular proper-ties of our material candidates and are hence most useful to assessmacroscopic quantities that primarily depend on them, such asthe Voc. The later stages will shift the focus to intermolecular andcondensed phase characteristics. The latter are central to, e.g.,exciton and charge transport processes,46,51,52,104�106 for whichmolecular properties alone are clearly of limited value. Bulkstructure considerations impact quantities such as the externalquantum efficiency and thus Jsc as well as the overall PCE. Sincethe cost and complexity of such studies increase significantly, theycan only be performed for a subset of the highest rated candidates.

In the first CEP phase, we perform a set of density functionaltheory (DFT) calculations108,109 employing the BP86,110,111

B3LYP,110,112,113 PBE0,114�119 BH&HLYP,110,112,120 and M06-2X121,122 functionals as well as the Hartree�Fock (HF) theory incombination with the single-ζ STO-6G,123,124 double-ζ def2-SVP, and triple-ζ def2-TZVP125 basis sets. We test and compareboth restricted and spin-polarized settings. Our selection offunctionals covers both generalized gradient approximation(GGA) and hybrid designs with a progression in the amountof exact exchange113 (BP86 and HF can be seen as the limitingcases). The latter has a systematic influence on orbital localiza-tion and thus eigenvalues.126 The GGA BP86 is a cost-effectiveway to obtain good geometries, B3LYP is arguably the mostwidely used functional in molecular quantum mechanics, andPBE0 has shown favorable performance in a variety of areas, ashas the highly parametrized M06-2X. (We will further test meta-GGAs such as TPSS127 and double-hybrids such as B2PLYP128

when they become available for CEP.) This range of modelchemistries was chosen to put our analysis on a broaderfooting, but also to assess the performance of the differenttheoretical methods.109 In total, each molecule is characterizedby a BP86/def2-SVP geometry optimization and 14 single-pointground state calculations. We obtain geometries, total energies(including their decomposition into different contributions),molecular orbitals (MOs) and their energy eigenvalues, electronand spin densities, electrostatic potentials, multipole moments,Mulliken129,130 and natural populations,131,132 as well as naturalatomic, localized molecular, and bond orbital analyses133�135 forthe different model chemistries.136

These basic electronic quantities can be used as a firstapproximation to the following points. The MO energies and

their differences can be related to ionization potentials, electronaffinities, gaps, and partial density of states (we note thatthe application of Koopmans’ theorem is problematic inDFT).126,137�142 The electronic levels have to be tuned for anefficient light absorption, for the necessary interplay betweendonor, acceptor, and lead materials, as well as for atmosphericstability.45 The delocalization of the frontier orbitals can beassociated with (intramolecular) exciton and charge carriermobility.143�147 Their spatial overlap indicates the transitioncharacter and probability of the corresponding excitation.148�150

Excess spin densities reflect the inadequacy of simple closed-shellsolutions and can also point to a complex and potentiallyunstable electronic situation.151 Charge maps can identify che-mically unstable sites in these highly unsaturated molecules aswell as patterns that may have an impact on their packing in thecondensed phase. The molecular multipole moments can corre-spondingly be correlated to the organization in the bulk structureand also to transport properties.152�155 The wave functionanalysis techniques are of interest for the study of structure�property relations. The obtained data can furthermore be utilizedas quantum chemical descriptors for the models described insection III. The results of a consecutive oligomer series can beused to extrapolate to the polymer limit.156�159

One example for how this basic information can be employedis the model developed by Scharber et al. in ref 107. The authorsassume (based on observations on actual OPV materials) thatcondensed matter aspects in such systems can be tuned to acertain viable level (i.e., a fill factor and external quantumefficiency of each 65% can be achieved). If that is the case, theirPCE can be approximated using only gap and lowest unoccupiedmolecular orbital (LUMO) energy values. Thismodel can readilybe applied to the current CEP data as shown in Figure 4. Ourpreliminary analysis reveals that only about 0.3% of the screenedcompounds (i.e., presently resulting in ∼3000�5000 distinctstructures, depending on the model chemistry) have the neces-sary energetic levels to realize OPVs with 10% or higherefficiency. This underscores the importance of carefully selectingthe compounds to be synthesized and tested, and at the sametime the value of the fast theoretical characterization and exten-sive search that CEP can provide toward this task. An unaidedsearch has only a small chance of success, while CEP finds severalthousand suitable structures. Figure 4 also displays the dynamicgap range and dipole moment distribution of the screened candi-dates as examples for the wide range of electronic propertiesfound in the CEP library. This versatility will be vital to asuccessful discovery of materials with specifically adjusted fea-tures (e.g., for different acceptors or tandem devices160).

We again emphasize that this first phase of CEP only addressesa subset of the important material issues161 and has the inherentaccuracy of the employed model chemistries and calibrations, iftaken at face value. There are, however, four factors that addconsiderable value to these calculations: (i) we correlate the com-puted results to actual experimental quantities to provide insightsinto their relationship; (ii) we use the electronic structure data asa source for new cheminformatics descriptors, which will put ourmodeling efforts on a more physical foundation; (iii) the analysisof the aggregated data frommillions of molecules in combinationwith structural similarity measures can reveal guiding trends,even if the absolute result for an individual candidate might beinaccurate due to a particular limitation of its electronic structure;(iv) by employing a variety of different model chemistries,we compensate for the chance of such a failure in any particular



method. We do not rely on any single result but use a compositescoring with many contributions.

The calculations allow for the elimination of unfit candidatesand provide a first set of predictions for promising structuraltrends. For these we can carry out the subsequent levels ofincreasingly more sophisticated calculations, which are plannedas follows: we will calculate (i) vibrational spectra and partitionfunctions to gauge phonon-scattering and trapping in vibrationalmodes;162,163 (ii) anionic/cationic states for improved electronaffinity/ionization potential values (a Dyson orbital analysis164

can improve our insights into the charge transfer processes);165

(iii) optimized geometries in the ionic states to deduce thereorganization during charge migration;163 (iv) triplet states andgaps to indicate potential for singlet fission processes;166 (v) linearresponse properties to assess the charge mobility;167,168 and (vi)excited states employing the maximum overlap method169 and/or(range-separated) time-dependent DFT170,171 with an electronattachment/detachment density172 and natural transition orbitalanalysis173 for a more sound description of the absorption process.Each higher level result can be used to assess the interpretations at thelower levels. Further studies could involve the calculation of transferintegrals between oligomer pairs,163 packing and interactions in thebulk phase, as well as the use of high-level wave function theory formore reliable results in complicated bonding situations. We also wantto consider the opposite approach, i.e., how well simple semiempiricaland model Hamiltonian calculations (which are popular in othercommunities) perform compared to first-principles DFT.101,174 Possi-ble stages of the CEP hierarchy were recently vetted in a successfulproof-of-principle study: we predicted exceptional charge mobility in anovel organic semiconductor, and this prediction could be confirmedexperimentally by the Bao Group with values of up to 16 cm2

V�1 s�1.78 We also used quantum chemical calculations in an earlierstudy to explain the observed mobility values in another system.175

V. Empirical Result Calibration. To bridge the gap betweentheory and experiment, we have introduced an empirical calibration

of the computational results. Such a calibration is a pragmatic wayto approximately account and correct for differences in experi-mental and theoretical property definitions, as well as in vacuoversus bulk, and oligomer versus polymer results.

We have established the organic electronics 2011 (OE11)training set of experimentally well-characterized organic electronicmaterials for the calibration and aligned the theoretical findingswith the corresponding data from experiment. The current calibra-tion is largely based on data from bulk-heterojunction setupswith PCBM as acceptor and introduces a corresponding bias. Adifferent focus can be chosen provided appropriate reference data.The use of training sets is a common technique in other areas ofquantum chemistry.121,176 The details of this work will be pre-sented elsewhere. A preliminary calibration of the current CEPresults was used in the analysis shown in Figure 4, and panels a�din particular demonstrate the success of this approach.

VI. Database and DataMining. The results of all calculations areused to build up a reference database: the Clean Energy ProjectDatabase (CEPDB). This data collection is comparable to themore general but much smaller NIST Computational ChemistryComparison and Benchmark Database (CCCBDB).177 As men-tioned before, the information accumulated in CEPDB is notonly relevant to OPVs but to organic electronics in general. Itis also designed to provide benchmarks for the performanceof various theoretical methods in this family of systems as well asa parameter repository for other calculations (e.g., for modelHamiltonians178�180 or custom force fields181). It will be avail-able to the public by 2012.

The primary purpose of CEPDB is to store and provide accessto the CEP data. Candidates with a certain set of desired para-meters can readily be identified. CEPDB also serves as the hubfor all data mining, analysis, and scoring operations to facilitatethe study of global trends, correlations, and OPV design rules.Finally, CEPDB is also responsible for bookkeeping, archiving theraw data (including the binary MO eigenvectors for subsequent

Figure 4. (a�e) Distribution of CEP hits in the gap�LUMOplane suggested by Scharber et al.107 Note that the density plots are on a logarithmic scale,and the red entries correspond toO (10 000) compounds andO (150 000) individual electronic structure calculations. A total of 2 600 000 compoundsare represented in these plots: (a,b) raw data from BP86/def2-SVP and BP86/def2-SVP//HF/def2-SVP (i.e., calculations which incorporate 0% vs100% exact exchange), respectively; (c,d) corresponding data after preliminary calibration; (e) OPV relevant parameter space with the 10% PCE region(with respect to a PCBM acceptor). About 0.3% of the screened compounds fall in this high-efficiency region. (f) Dynamic gap range, i.e., the range ofavailable LUMO energies given a particular HOMO and vice versa; (g) PCE histogram according to the Scharber model; (h) dipole moment histogram.Panels e�h are all at the BP86/def2-SVP//PBE0/def2-TZVP level of theory (calibrated).



calculations), keeping track of the project progress, and prioritiz-ing the study, which are clearly important tasks considering thescope of the project and the volume of data.

VII. The World Community Grid. The massive demand incomputing time for CEP is largely provided by the WCG,182,183

a distributed volunteer computing platform for philanthropicresearch organized by IBM. Presently, ∼560 000 users havesigned up ∼1 800 000 computers to the various WCG projects.Participants can donate computing time by running the sup-ported science applications on their personal computers, eitheron low priority in the background or in screensaver mode duringidle times (see Figure 5). In CEP, we use a custom version of theQ-Chem3.2 programpackage,184 whichwas ported to the BerkeleyOpen Infrastructure for Network Computing (BOINC)185

environment. From the user perspective, the participation in aWCG project is fully automated and usually does not require anyinput or maintenance beyond the initial setup.

The WCG is a powerful resource and provides us with themeans for our high-throughput investigation. It is, however, alsounusual due to the nonspecialized hardware and host demands.These limitations have to be taken into consideration in thedesign of suitable tasks. In addition to theWCG, CEP utilizes theHarvard FAS Odyssey cluster as well as accounts at NERSC186

and TeraGrid187 for problems that are outside the scope ofvolunteer grid computing (e.g., due to their size and computa-tional complexity). CEP currently characterizes about 20 000oligomer sequences per day, and so far we have studied∼2 600 000 structures in ∼36 000 000 calculations, utilizing∼4500 years of CPU time. Close to 100% of these calculationswere performed on the grid, but the use of cluster resources willbecome more important in the more demanding stages of the

project. Their contribution in terms of volume, however, willremain very limited.

One important aspect in a computational study at this scale isthat all processes have to be automated to keep it feasible,efficient, and reduce human error. We created the necessaryfacilities using the Python scripting language.188,189

This initial presentation of theHarvardClean Energy Project hasoutlined its overall architecture, the machinery we put in place, andupcoming extensions. We pointed to the first results, and moredetailed reports on the various aspects of the project—tied togetherby this Perspective—will be given in subsequent publications.

CEP applies a modern cyberinfrastructure paradigm to com-putational materials science and, in particular, to renewableenergy research. Engineering successful OPV materials is acomplex challenge as a number of optical and electronic require-ments have to be met. We showed that CEP with its large-scalescreening is well equipped for a knowledge-based search ofsystems with suitable features. It provides a unique access todata for a wide array of potential compounds with diverseelectronic structures, and it is ideally suited to identify highlypromising donor or acceptor candidates in the infinite space oforganic electronics. Our work can thus complement and accel-erate traditional research approaches, and it can help develop anunderstanding of structure�property relationships to facilitatethe rational design of new materials. We hope that joint effortswith experimental collaborators can contribute to overcomingthe current limitations of OPV materials in order to provide aclean source of electricity that can compete with conventionalpower production.

’AUTHOR INFORMATION

Corresponding Author*Electronic address: [email protected] (J.H.); [email protected] (A.A.-G.).

Present Addresses

)Research and Technology Center North America, Robert BoschLLC, Cambridge, MA 02142.

’BIOGRAPHIES

Johannes Hachmann obtained his Dipl.-Chem. degree(2004) after undergraduate studies at the University of Jena,Germany, and the University of Cambridge, U.K. He received anM.Sc. (2007) and Ph.D. (2010) from Cornell University underthe supervision of Prof. G.K.-L. Chan working on density matrixrenormalization group theory and computational transitionmetal chemistry. In 2009 he joined the Aspuru-Guzik Groupand the Clean Energy Project at Harvard University.

Figure 5. Top: CEP project homepage and hub for project participants.Bottom: The CEP “screensaver” as it is displayed on volunteer hosts.

CEP applies a modern cyberinfras-tructure paradigm to computa-tional materials science and, inparticular, to renewable energy

research.



RobertoOlivares-Amaya is a Ph.D. Chemical Physics studentin the Aspuru-Guzik Group at Harvard University and was a2010�2011 Giorgio Ruffolo Sustainability Science DoctoralFellow. He obtained his B.Sc. degree (2007) from the NationalAutonomous University of Mexico (UNAM). His researchinterests include the study of structure and response of OPVmaterials by means of first-principles electronic structure andmachine learning methods.

Sule Atahan-Evrenk is a postdoctoral research fellow in theAspuru-Guzik Group at Harvard University. She received a B.S.degree in chemistry from Bilkent University, Turkey, and a Ph.D.in physical chemistry from the University of Maryland CollegePark. Her research interests include the theoretical characteriza-tion of organic electronic materials by quantum chemistry andmolecular dynamics methods.

Carlos Amador-Bedolla is a professor at the Facultad deQuímica of the National Autonomous University of Mexico inMexico City. He obtained his doctorate in chemistry at UNAM(1989) and has worked with several leading groups at CaseWestern Reserve University, UC Berkeley, and Harvard Uni-versity. His current research focuses on renewable energymaterials (see www.quimica.unam.mx/ficha_investigador.php?ID=77&tipo=2).

Roel S. S�anchez-Carrera is currently a research staff at theRobert Bosch Research and Technology Center North America.He obtained his B.S. in chemistry from theMonterrey Institute ofTechnology and Higher Education, M�exico, and a Ph.D. inphysical chemistry from the Georgia Institute of Technologyunder the supervision of Prof. J.-L. Br�edas. He then moved toHarvard University, where he worked in the group of Prof. Aspuru-Guzik as a Mary-Fieser Postdoctoral Fellow.

Aryeh Gold-Parker is an undergraduate student at HarvardUniversity and will complete his B.A. degree in chemistry andphysics in 2012. His research interests are in physical chemistry,specifically in renewable energy materials. He joined theAspuru-Guzik Group in 2009 to work on the Clean EnergyProject.

Leslie Vogt completed a B.S. in physical chemistry at JuniataCollege in 2005 and spent 2006�2007 working with Prof. R.Zimmermann as a Fulbright Fellow in Germany. She is complet-ing her Ph.D. at Harvard University in the Aspuru-Guzik Groupstudying the electronic properties of organic materials as an NSFGraduate Research Fellow.

AnnaM. Brockway is a chemistry senior at Haverford Collegewhere she works with Prof. J. Schrier in the field of computationaltheoretical chemistry. She joined the Clean Energy Project in2010 during her time as an REU summer student at HarvardUniversity.

Al�an Aspuru-Guzik holds a Ph.D. in physical chemistry fromUC Berkeley, and is currently an associate professor of Chem-istry and Chemical Biology at Harvard University. His researchlies at the intersection of quantum information/computation andtheoretical chemistry. He is interested in energy transfer dynamicsand renewable energy materials. (See http://aspuru.chem.harvard.edu/ and the project webpage http://cleanenergy.harvard.edu/.)

’ACKNOWLEDGMENT

We wish to thank IBM for organizing the World CommunityGrid and the WCG members for their computing time dona-tions. Additional information about the project as well as links tojoin the WCG and CEP can be found in refs 44 and 182. We are

grateful to our experimental collaborators Zhenan Bao, RajibMondal, and Anatoliy N. Sokolov at Stanford University; forsupport from the Q-Chem development team by Yihan Shao,Alex J. Sodt, and Zhengting Gan; for support from the IBMWCG team by Jonathan D. Armstrong, Kevin N. Reed, Keith J.Uplinger, Al Seippel, Viktors Berstis, and Bill Bovermann; forsupport from Harvard FAS Research Computing by SuvendraDutta, Christopher Walker, Gerald I. Lotto, and James Cuff; andto Aidan Daly, Erica Lin, and Lauren A. Kaye at Harvard. TheCEP is supported by the National Science Foundation throughGrant Nos. DMR-0820484 and DMR-0934480 as well asthe Global Climate and Energy Project (Stanford Grant No.25591130-45282-A. Any opinions, findings, conclusions, orrecommendations expressed in this publication are those of theauthor(s) and do not necessarily reflect the views of StanfordUniversity, the Sponsors of the Global Climate and EnergyProject, or others involved with the Global Climate and EnergyProject). R.O.-A. was supported by a Giorgio Ruffolo Fellowshipin the Sustainability Science Program at Harvard University’sCenter for International Development, C.A.-B. acknowledgesfinancial support from CONACyT M�exico and computer accessto DGTIC UNAM, A.G.-P. acknowledges funding by theHarvard College Research Program and the Harvard UniversityCenter for the Environment Summer Research Fund, R.S.S.-C.thanks the Mary Fieser Postdoctoral Fellowship at HarvardUniversity, L.V. was supported by an NSF Graduate ResearchFellowship, and A.M.B. was supported by the Harvard REUprogram. We acknowledge software support from Q-Chem Inc.,Optibrium Ltd., Dotmatics Ltd., Molecular Networks GmbH,and ChemAxon Ltd. Additional computing time was provided onthe Odyssey cluster supported by the FAS Research ComputingGroup, NERSC, and the National Science Foundation TeraGrid.

’REFERENCES

(1) W€ohrle, D.; Meissner, D. Organic Solar Cells. Adv. Mater. 1991,3, 129–138.

(2) Conti, J. J.; Holtberg, P. D.; Beamon, J. A.; Schaal, A. M.; Ayoub,J. C.; Turnure, J. T. Annual Energy Outlook 2011; U.S. Energy Informa-tion Administration: Washington, DC, 2011.

(3) Zweibel, K.; Mason, J.; Fthenakis, V. A Solar Grand Plan. Sci. Am.2008, 298, 64–73.

(4) Jacobson, M. Z.; Delucchi, M. A. A Plan to Power 100% of thePlanet with Renewables. Sci. Am. 2009, 301, 58–65.

(5) Lewis, N. S. Toward Cost-Effective Solar Energy Use. Science2007, 315, 798–801.

(6) Boxwell, M. Solar Electricity Handbook � 2011 Edition, 4th ed.;Greenstream Publishing: Ryton on Dunsmore, U.K., 2011.

(7) Brabec, C. J.; Dyakonov, V.; Parisi, J.; Sariciftci, N. S. OrganicPhotovoltaics: Concepts and Realization; Springer: Berlin, 2010.

(8) Brabec, C. J.; Dyakonov, V.; Scherf, U. Organic Photovoltaics:Materials, Device Physics, and Manufacturing Technologies; Wiley-VCH:Weinheim, Germany, 2008.

(9) Sun, S.-S.; Sariciftci, N. S. Organic Photovoltaics: Mechanisms,Materials, and Devices; CRC Press: Boca Raton, FL, 2005.

(10) Tang, C. W. Two-Layer Organic Photovoltaic Cell. Appl. Phys.Lett. 1986, 48, 183–185.

(11) Beaumont, N.; Hancox, I.; Sullivan, P.; Hatton, R. A.; Jones,T. S. Increased Efficiency in Small Molecule Organic Photovoltaic Cellsthrough Electrode Modification with Self-Assembled Monolayers.Energy Environ. Sci. 2011, 4, 1708–1711.

(12) Riede, M.; Mueller, T.; Tress, W.; Schueppel, R.; Leo, K. Small-Molecule Solar Cells � Status and Perspectives. Nanotechnology 2008,19, 424001.



(13) Kalyanasundaram, K. Dye-Sensitized Solar Cells; CRC Press:Boca Raton, FL, 2010.(14) Krebs, F. C. Polymer Photovoltaics: A Practical Approach; SPIE

Publications: Bellingham, WA, 2008.(15) Pagliaro, M.; Palmisano, G.; Ciriminna, R. Flexible Solar Cells;

Wiley-VCH: Weinheim, Germany, 2008.(16) Green, M. A. Third Generation Photovoltaics: Solar Cells for

2020 and Beyond. Physica E 2002, 14, 65–70.(17) Galagan, Y.; de Vries, I. G.; Langen, A. P.; Andriessen, R.;

Verhees, W. J.; Veenstra, S. C.; Kroon, J. M. Technology Developmentfor Roll-to-Roll Production of Organic Photovoltaics. Chem. Eng. Process2011, 50, 454–461.(18) Krebs, F. C.; Tromholt, T.; Jørgensen, M. Upscaling of Polymer

Solar Cell Fabrication Using Full Roll-to-Roll Processing. Nanoscale2010, 2, 873–886.(19) Vak, D.; Kim, S.; Jo, J.; Oh, S.; Na, S.; Kim, J.; Kim, D.

Fabrication of Organic Bulk Heterojunction Solar Cells by a SprayDeposition Method for Low-Cost Power Generation. Appl. Phys. Lett.2007, 91, 081102.(20) Green, R.; Morpha, A.; Ferguson, A. J.; Kopidakis, N.; Rumbles,

G.; Shaheen, S. E. Performance of Bulk Heterojunction PhotovoltaicDevices Prepared by Airbrush Spray Deposition. Appl. Phys. Lett. 2008,92, 33301.(21) Steirer, K. X.; Reese, M. O.; Rupert, B. L.; Kopidakis, N.; Olson,

D. C.; Collins, R. T.; Ginley, D. S. Ultrasonic Spray Deposition forProduction of Organic Solar Cells. Sol. Energy Mater. Sol. Cells 2009, 93,447–453.(22) O’Neill, M.; Kelly, S. M. Ordered Materials for Organic

Electronics and Photonics. Adv. Mater. 2011, 23, 566–584.(23) Bailey-Salzman, R. F.; Rand, B. P.; Forrest, S. R. Semitranspar-

ent Organic Photovoltaic Cells. Appl. Phys. Lett. 2006, 88, 233502.(24) Colsmann, A.; Puetz, A.; Bauer, A.; Hanisch, J.; Ahlswede, E.;

Lemmer, U. Efficient Semi-Transparent Organic Solar Cells with GoodTransparency Color Perception and Rendering Properties. Adv. EnergyMater. 2011, 1, 599–603.(25) Pagliaro, M.; Ciriminna, R.; Palmisano, G. Flexible Solar Cells.

ChemSusChem 2008, 1, 880–891.(26) Heeger, A. InGlobal Sustainability - ANobel Cause; Schellnhuber,

H. J., Molina, M., Stern, N., Huber, V., Kadner, S., Eds.; CambridgeUniversity Press: Cambridge, U.K., 2010.(27) Forrest, S. R. The Path to Ubiquitous and Low-Cost Organic

Electronic Appliances on Plastic. Nature 2004, 428, 911–918.(28) Marsh, G. BIPV: Innovation Puts Spotlight on Solar. Renewable

Energy Focus 2008, 9, 62–67.(29) Henemann, A. BIPV: Built-In Solar Energy. Renewable Energy

Focus 2008, 9 (14), 16–19.(30) Tipnis, R.; Bernkopf, J.; Jia, S.; Krieg, J.; Li, S.; Storch, M.; Laird,

D. Large-Area Organic Photovoltaic Module � Fabrication and Perfor-mance. Sol. Energy Mater. Sol. Cells 2009, 93, 442–446.(31) Schubert, M. B.; Werner, J. H. Flexible Solar Cells for Clothing.

Mater. Today 2006, 9, 42–50.(32) Brabec, C. J. Organic Photovoltaics: Technology and Market.

Sol. Energy Mater. Sol. Cells 2004, 83, 273–292.(33) Bedeloglu, A.; Demir, A.; Bozkurt, Y.; Sariciftci, N. S. A Flexible

Textile Structure Based on Polymeric Photovoltaics Using TransparentCathode. Synth. Met. 2009, 159, 2043–2048.(34) http://wgsi.org/files/EquinoxCommunique_June9_2011.pdf(35) Shrotriya, V. Organic Photovoltaics: Polymer Power. Nat.

Photonics 2009, 3, 447–449.(36) Peters, C. H.; Sachs-Quitana, I. T.; Kastrop, J. P.; Beaupr�e, S.;

Leclerc, M.; McGehee, M. D. High Efficiency Polymer Solar Cells withLong Operating Lifetimes. Adv. Energy Mater. 2011, 1, 491–494.(37) Service, R. F. Outlook Brightens for Plastic Solar Cells. Science

2011, 332, 293.(38) Rivaton, A.; Chambon, S.; Manceau, M.; Gardette, J.-L.;

Lemaître, N.; Guillerez, S. Light-Induced Degradation of the ActiveLayer of Polymer-Based Solar Cells. Polym. Degrad. Stab. 2010, 95,278–284.

(39) Manceau,M.; Rivaton, A.; Gardette, J.-L.; Guillerez, S.; Lemaître,N. Light-Induced Degradation of the P3HT-Based Solar Cells ActiveLayer. Sol. Energy Mater. Sol. Cells 2011, 95, 1315–1325.

(40) Bertho, S.; Haeldermans, I.; Swinnen, A.; Moons, W.; Martens,T.; Lutsen, L.; Vanderzande, D.; Manca, J.; Senes, A.; Bonfiglio, A.Influence of Thermal Ageing on the Stability of Polymer Bulk Hetero-junction Solar Cells. Sol. Energy Mater. Sol. Cells 2007, 91, 385–389.

(41) Bertho, S.; Janssen, G.; Cleij, T. J.; Conings, B.; Moons, W.;Gadisa, A.; D'Haen, J.; Goovaerts, E.; Lutsen, L.;Manca, J.; et al. Effect ofTemperature on the Morphological and Photovoltaic Stability of BulkHeterojunction Polymer:Fullerene Solar Cells. Sol. Energy Mater. Sol.Cells 2008, 92, 753–760.

(42) Gaudiana, R.; Brabec, C. Organic Materials: Fantastic Plastic.Nat. Photonics 2008, 2, 287–289.

(43) Shaheen, S. E.; Ginley, D. S.; Jabbour, G. E. Organic-Based Photo-voltaics: Toward Low-Cost Power Generation.MRS Bull. 2005, 30, 10–15.

(44) Harvard CEP Home Page. http://cleanenergy.harvard.edu(accessed August 10, 2011).

(45) Br�edas, J.-L.; Norton, J. E.; Cornil, J.; Coropceanu, V.MolecularUnderstanding of Organic Solar Cells: The Challenges. Acc. Chem. Res.2009, 42, 1691–1699.

(46) Heremans, P.; Cheyns, D.; Rand, B. P. Strategies for Increasingthe Efficiency of Heterojunction Organic Solar Cells: Material Selectionand Device Architecture. Acc. Chem. Res. 2009, 42, 1740–1747.

(47) Chen, J.; Cao, Y. Development of Novel Conjugated DonorPolymers forHigh-Efficiency Bulk-Heterojunction Photovoltaic Devices.Acc. Chem. Res. 2009, 42, 1709–1718.

(48) Schlenker, C. W.; Thompson, M. E. The Molecular Nature ofPhotovoltage Losses in Organic Solar Cells. Chem. Commun. 2011, 47,3702–3716.

(49) Borchert, H. Elementary Processes and Limiting Factors inHybrid Polymer/Nanoparticle Solar Cells. Energy Environ. Sci. 2010, 3,1682–1694.

(50) Blouin, N.; Michaud, A.; Gendron, D.; Wakim, S.; Blair, E.;Neagu-Plesu, R.; Belletete, M.; Durocher, G.; Tao, Y.; Leclerc, M.Toward a Rational Design of Poly(2,7-carbazole) Derivatives for SolarCells. J. Am. Chem. Soc. 2008, 130, 732–742.

(51) Nelson, J.; Kwiatkowski, J. J.; Kirkpatrick, J.; Frost, J. M.Modeling Charge Transport in Organic Photovoltaic Materials. Acc.Chem. Res. 2009, 42, 1768–1778.

(52) Zhu, X.-Y.; Yang, Q.; Muntwiler, M. Charge-Transfer Excitonsat Organic Semiconductor Surfaces and Interfaces. Acc. Chem. Res. 2009,42, 1779–1787.

(53) Wang, L.; Nan, G.; Yang, X.; Peng, Q.; Li, Q.; Shuai, Z. Compu-tational Methods for Design of Organic Materials with High ChargeMobility. Chem. Soc. Rev. 2010, 39, 423–434.

(54) Kerns, E. H. High Throughput Physicochemical Profiling forDrug Discovery. J. Pharm. Sci. 2001, 90, 1838–1858.

(55) Bajorath, J. Integration of Virtual and High-ThroughputScreening. Nat. Rev. Drug Discovery 2002, 1, 882–894.

(56) McInnes, C. Virtual Screening Strategies in Drug Discovery.Curr. Opin. Chem. Biol. 2007, 11, 494–502.

(57) Gribbon, P.; Andreas, S. High-Throughput Drug Discovery:What Can We Expect from HTS?. Drug Discovery Today 2005, 10,17–22.

(58) Appell, K.; Baldwin, J. J.; Egan, W. J. In Handbook of ModernPharmaceutical Analysis: Separation Science and Technology; Ahuja, S.,Scypinski, S., Eds.; Academic Press: San Diego, CA, 2001; Vol. 3, pp23�56.

(59) Gordon, E. M.; Gallop, M. A.; Patel, D. V. Strategy and Tacticsin Combinatorial Organic Synthesis. Applications to Drug Discovery.Acc. Chem. Res. 1996, 29, 144–154.

(60) Ellman, J. A. Design, Synthesis, and Evaluation of Small-Molecule Libraries. Acc. Chem. Res. 1996, 29, 132–143.

(61) Gordon, E. M.; Barrett, R. W.; Dower, W. J.; Fodor, S. P. A.;Gallop, M. A. Applications of Combinatorial Technologies to Drug Dis-covery. 2. Combinatorial Organic Synthesis, Library Screening Strate-gies, and Future Directions. J. Med. Chem. 1994, 37, 1385–1401.



(62) The authors are aware of and acknowledge related work byGeoffreyHutchison (University of Pittsburgh), Michel Cot�e (Universityof Montreal), and Josef Michl (University of Colorado). Their respec-tive research efforts each have a certain degree of overlap with CEP.(63) Bajorath, J. Chemoinformatics: Concepts, Methods, and Tools for

Drug Discovery; Humana Press: Totowa, NJ, 2010.(64) Ghose, A. K.; Herbertz, T.; Salvino, J. M.; Mallamo, J. P.

Knowledge-Based Chemoinformatic Approaches to Drug Discovery.Drug Discovery Today 2006, 11, 1107–1114.(65) Gaber, M. M. Scientific Data Mining and Knowledge Discovery:

Principles and Foundations; Springer: Berlin, 2009.(66) Grossman, R. L.; Kamath, C.; Kegelmeyer, P.; Kumar, V.;

Namburu, R. Data Mining for Scientific and Engineering Applications;Springer: Berlin, 2001.(67) Suh, C.; Rajan, K. Invited Review: DataMining and Informatics

for Crystal Chemistry: Establishing Measurement Techniques for Map-ping Structure-Property Relationships. Mater. Sci. Technol. 2009, 25,466–471.(68) Rajan, K. Combinatorial Materials Sciences: Experimental

Strategies for Accelerated Knowledge Discovery. Annu. Rev. Mater. Res.2008, 38, 299–322.(69) Suh, C.; Rajan, K. Virtual Screening and QSAR Formulations

for Crystal Chemistry. QSAR Comb. Sci. 2005, 24, 114–119.(70) Jain, A.; Hautier, G.; Moore, C. J.; Ong, S. P.; Fischer, C. C.;

Mueller, T.; Persson, K. A.; Ceder, G. AHigh-Throughput Infrastructurefor Density Functional Theory Calculations. Comput. Mater. Sci. 2011,50, 2295–2310.(71) Colella, S.; Mazzeo, M.; Grisorio, R.; Fabiano, E.; Melcarne, G.;

Carallo, S.; Angione, M. D.; Torsi, L.; Suranna, G. P.; della Sala, F.; et al.MonodispersedMolecular Donors for Bulk Hetero-junction Solar Cells:From Molecular Properties to Device Performances. Chem. Commun.2010, 46, 6273–6275.(72) Zhan, X.; Zhu, D. Conjugated Polymers for High-Efficiency

Organic Photovoltaics. Polym. Chem. 2010, 1, 409–419.(73) Liang, Y.; Feng, D.; Wu, Y.; Tsai, S.-T.; Li, G.; Ray, C.; Yu, L.

Highly Efficient Solar Cell Polymers Developed via Fine-Tuning of Structuraland Electronic Properties. J. Am. Chem. Soc. 2009, 131, 7792–7799.(74) Chochos, C. L.; Choulis, S. A. How the Structural Deviations

on the Backbone of Conjugated Polymers Influence Their Optoelec-tronic Properties and Photovoltaic Performance. Prog. Polym. Sci. 2011,36, 1326–1414.(75) Liu, J.; Mikhaylov, I. A.; Zou, J.; Osaka, I.; Masunov, A. E.;

McCullough, R. D.; Zhai, L. Insight into How Molecular Structures ofThiophene-Based Conjugated Polymers Affect Crystallization Beha-viors. Polymer 2011, 52, 2302–2309.(76) Mishra, A.; Fischer, M. K. R.; B€auerle, P. Metal-Free Organic

Dyes for Dye-Sensitized Solar Cells: From Structure:Property Relation-ships to Design Rules. Angew. Chem., Int. Ed. 2009, 48, 2474–2499.(77) Beaujuge, P. M.; Amb, C. M.; Reynolds, J. R. Spectral En-

gineering in π-Conjugated Polymers with Intramolecular Donor�Acceptor Interactions. Acc. Chem. Res. 2010, 43, 1396–1407.(78) Sokolov, A. N.; Atahan-Evrenk, S.; Mondal, R.; Akkerman,

H. B.; S�anchez-Carrera, R. S.; Granados-Focil, S.; Schrier, J.; Mannsfeld,S. C. B.; Zoombelt, A. P.; Bao, Z.; Aspuru-Guzik, A. From Computa-tional Discovery to Experimental Characterization of a High HoleMobility Organic Crystal. Nat. Commun. 2011, 2, 437.(79) Facchetti, A. π-Conjugated Polymers for Organic Electronics

and Photovoltaic Cell Applications. Chem. Mater. 2011, 23, 733–758.(80) Pron, A.; Rannou, P. Processible Conjugated Polymers: From

Organic Semiconductors to Organic Metals and Superconductors. Prog.Polym. Sci. 2002, 27, 135–190.(81) Weininger, D. SMILES, A Chemical Language and Information

System. 1. Introduction to Methodology and Encoding Rules. J. Chem.Inf. Comput. Sci. 1988, 28, 31–36.(82) Marvin 5.3.2 by ChemAxon, 2010, http://www.chemaxon.com(83) Sadowski, J.; Gasteiger, J.; Klebe, G. Comparison of Automatic

Three-DimensionalModel Builders Using 639 X-ray Structures. J. Chem.Inf. Comput. Sci. 1994, 34, 1000–1008.

(84) Farchioni, R.; Grosso, G. Organic Electronic Materials: Conju-gated Polymers and Low Molecular Weight Organic Solids; Springer:Berlin, 2001.

(85) Meller, G.; Grasser, T. Organic Electronics; Springer: Berlin,2009.

(86) Paszkowicz, W. Genetic Algorithms, A Nature-Inspired Tool:Survey of Applications in Materials Science and Related Fields. Mater.Manuf. Processes 2009, 24, 174–197.

(87) Arora, V.; Bakhshi, A. Molecular Designing of Novel TernaryCopolymers of Donor�Acceptor Polymers Using Genetic Algorithm.Chem. Phys. 2010, 373, 307–312.

(88) Giro, R.; Cyrillo, M.; Galv~ao, D. S. Designing ConductingPolymers Using Genetic Algorithms. Chem. Phys. Lett. 2002, 366,170–175.

(89) Martins, B. V. C.; Brunetto, G.; Sato, F.; Coluci, V. R.; Galv~ao,D. S. Designing Conducting Polymers Using Bioinspired Ant Algo-rithms. Chem. Phys. Lett. 2008, 453, 290–295.

(90) Lameijer, E.-W.; B€ack, T.; Kok, J.; Ijzerman, A. D. EvolutionaryAlgorithms in Drug Design. Nat. Comput. 2005, 4, 177–243.

(91) Todeschini, R.; Consonni, V. Handbook of Molecular Descrip-tors; Wiley-VCH: Weinheim, germany, 2002.

(92) Sykora, V. J.; Leahy, D. E. Chemical Descriptors Library(CDL): A Generic, Open Source Software Library for ChemicalInformatics. J. Chem. Inf. Model. 2008, 48, 1931–1942.

(93) Mitra, S.; Datta, S.; Perkins, T.; Michailidis, G. Introduction toMachine Learning and Bioinformatics; CRC Press: Boca Raton, FL, 2008.

(94) Yang, Z. R. Machine Learning Approaches to Bioinformatics;World Scientific: Singapore, 2010.

(95) Bishop, C. M. Pattern Recognition and Machine Learning, 2nded.; Springer: Berlin, 2007.

(96) Fukunaga, K. Introduction to Statistical Pattern Recognition, 2nded.; Academic Press: San Diego, CA, 1990.

(97) Olivares-Amaya, R.; Amador-Bedolla, C.; Hachmann, J.; Ata-han-Evrenk, S.; S�anchez-Carrera, R. S.; Vogt, L.; Aspuru-Guzik, A.Accelerated Computational Discovery of High-Performance Materialsfor Organic Photovoltaics by Means of Cheminformatics. EnergyEnviron. Sci. 2011, accepted for publication.

(98) RDevelopment Core Team. R: A Language and Environment forStatistical Computing; R Foundation for Statistical Computing: Vienna,Austria, 2008.

(99) Dragon 6 by Talete, 2011, http://www.talete.mi.it/index.htm(100) StarDrop 5.0 by Optibrium, 2011, http://www.optibrium.

com/stardrop/(101) Andr�e, J.-M.; Delhalle, J.; Br�edas, J.-L. Quantum Chemistry

Aided Design of Organic Polymers: An Introduction to the QuantumChemistry of Polymers and Its Applications; World Scientific: Singapore,1991.

(102) Risko, C.; McGehee, M. D.; Br�edas, J.-L. A Quantum-Chemi-cal Perspective into Low Optical-Gap Polymers for Highly-EfficientOrganic Solar Cells. Chem. Sci. 2011, 2, 1200–1218.

(103) Wang, L.; Li, Q.; Shuai, Z.; Chen, L.; Shi, Q. Multiscale Studyof Charge Mobility of Organic Semiconductor with Dynamic Disorders.Phys. Chem. Chem. Phys. 2010, 12, 3309–3314.

(104) Benanti, T.; Venkataraman, D. Organic Solar Cells: An Over-view Focusing on Active Layer Morphology. Photosynth. Res. 2006, 87,73–81.

(105) Gregg, B. A. In Molecules as Components of ElectronicDevices; Marya, L., Ed.; ACS: Washington, DC, 2003; Chapter 19,pp 243�257.

(106) Xiao, S.; Price, S. C.; Zhou, H.; You, W. In Functional PolymerNanocomposites for Energy Storage and Conversion; Qing,W., Lei, Z., Eds.;ACS: Washington, DC, 2010; Chapter 7, pp 71�80.

(107) Scharber, M. C.; M€uhlbacher, D.; Koppe, M.; Denk, P.;Heeger, A. J.; Waldauf, C.; Brabec, C. J. Design Rules for Donors inBulk-Heterojunction Solar Cells � Towards 10% Energy-ConversionEfficiency. Adv. Mater. 2006, 18, 789–794.

(108) Parr, R.; Yang, W. Density-Functional Theory of Atoms andMolecules; Oxford University Press: New York, 1989; p 333.



(109) Koch, W.; Holthausen, M. C. A Chemist’s Guide to DensityFunctional Theory, 2nd ed.; Wiley-VCH: Weinheim, Germany, 2001.(110) Becke, A. D. Density-Functional Exchange-Energy Approxima-

tion with Correct Asymptotic Behavior. Phys. Rev. A 1988, 38, 3098–3100.(111) Perdew, J. P. Density-Functional Approximation for the

Correlation Energy of the Inhomogeneous Electron Gas. Phys. Rev. B1986, 33, 8822–8824.(112) Lee, C.; Yang, W.; Parr, R. G. Development of the Colle�

Salvetti Correlation-Energy Formula into a Functional of the ElectronDensity. Phys. Rev. B 1988, 37, 785–789.(113) Becke, A. D. Density-Functional Thermochemistry. III. The

Role of Exact Exchange. J. Chem. Phys. 1993, 98, 5648–5652.(114) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient

Approximation Made Simple. Phys. Rev. Lett. 1996, 77, 3865–3868.(115) Perdew, J. P.; Ernzerhof, M.; Burke, K. Rationale for Mixing

Exact Exchange with Density Functional Approximations. J. Chem. Phys.1996, 105, 9982–9985.(116) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient

Approximation Made Simple [Phys. Rev. Lett. 77, 3865 (1996)]. Phys.Rev. Lett. 1997, 78, 1396.(117) Tao, J.; Perdew, J. P.; Staroverov, V. N.; Scuseria, G. E.

Climbing the Density Functional Ladder: Nonempirical Meta-General-ized Gradient Approximation Designed for Molecules and Solids. Phys.Rev. Lett. 2003, 91, 146401.(118) Perdew, J. P.; Tao, J.; Staroverov, V. N.; Scuseria, G. E. Meta-

Generalized Gradient Approximation: Explanation of a Realistic Non-empirical Density Functional. J. Chem. Phys. 2004, 120, 6898–6911.(119) Adamo, C.; Barone, V. Toward Reliable Density Functional

Methods without Adjustable Parameters: The PBE0 Model. J. Chem.Phys. 1999, 110, 6158–6169.(120) Becke, A. D. A New Mixing of Hartree�Fock and Local

Density-Functional Theories. J. Chem. Phys. 1993, 98, 1372–1377.(121) Zhao, Y.; Truhlar, D. G. The M06 Suite of Density Func-

tionals for Main Group Thermochemistry, Thermochemical Kinetics,Noncovalent interactions, Excited States, and Transition Elements: TwoNew Functionals and Systematic Testing of Four M06 Functionals andTwelve Other Functionals. Theor. Chem. Acc. 2008, 120, 215–241.(122) Zhao, Y.; Truhlar, D. G. Density Functionals with Broad

Applicability in Chemistry. Acc. Chem. Res. 2008, 41, 157–167.(123) Hehre, W. J.; Stewart, R. F.; Pople, J. A. Self-Consistent

Molecular-Orbital Methods. I. Use of Gaussian Expansions of Slater-Type Atomic Orbitals. J. Chem. Phys. 1969, 51, 2657–2664.(124) Hehre, W. J.; Ditchfield, R.; Stewart, R. F.; Pople, J. A. Self-

Consistent Molecular Orbital Methods. IV. Use of Gaussian Expansionsof Slater-Type Orbitals. Extension to Second-Row Molecules. J. Chem.Phys. 1970, 52, 2769–2773.(125) Weigend, F.; Ahlrichs, R. Balanced Basis Sets of Split Valence,

Triple Zeta Valence and Quadruple Zeta Valence Quality for H to Rn:Design and Assessment of Accuracy. Phys. Chem. Chem. Phys. 2005, 7,3297–3305.(126) Zhang, G.; Musgrave, C. B. Comparison of DFTMethods for

Molecular Orbital Eigenvalue Calculations. J. Phys. Chem. A 2007, 111,1554–1561.(127) Tao, J.; Perdew, J. P.; Staroverov, V. N.; Scuseria, G. E. Climb-

ing the Density Functional Ladder: Nonempirical Meta�GeneralizedGradient Approximation Designed for Molecules and Solids. Phys. Rev.Lett. 2003, 91, 146401.(128) Grimme, S. Semiempirical Hybrid Density Functional with

Perturbative Second-Order Correlation. J. Chem. Phys. 2006, 124, 034108.(129) Mulliken, R. S. Electronic Population Analysis on LCAO-MO

Molecular Wave Functions. I. J. Chem. Phys. 1955, 23, 1833–1840.(130) Wiberg, K. B.; Rablen, P. R. Comparison of Atomic Charges

Derived via Different Procedures. J. Comput. Chem. 1993, 14, 1504–1518.(131) Reed, A. E.; Weinhold, F. Natural Bond Orbital Analysis

of near-Hartree�Fock Water Dimer. J. Chem. Phys. 1983, 78, 4066–4073.(132) Reed, A. E.; Weinstock, R. B.; Weinhold, F. Natural Popula-

tion Analysis. J. Chem. Phys. 1985, 83, 735–746.

(133) Reed, A. E.; Weinhold, F. Natural Localized MolecularOrbitals. J. Chem. Phys. 1985, 83, 1736–1740.

(134) Foster, J. P.;Weinhold, F. Natural hybrid orbitals. J. Am. Chem.Soc. 1980, 102, 7211–7218.

(135) Reed, A. E.; Curtiss, L. A.; Weinhold, F. IntermolecularInteractions from a Natural Bond Orbital, Donor�Acceptor Viewpoint.Chem. Rev. 1988, 88, 899–926.

(136) Foresman, J. B.; Frisch, A. Exploring Chemistry with ElectronicStructure Methods, 2nd ed.; Gaussian Inc.: Pittsburgh, PA, 1996.

(137) Stowasser, R.; Hoffmann, R. What Do the Kohn�ShamOrbitals and Eigenvalues Mean?. J. Am. Chem. Soc. 1999, 121, 3414–3420.

(138) Hamel, S.; Duffy, P.; Casida, M. E.; Salahub, D. R. Kohn�Sham Orbitals and Orbital Energies: Fictitious Constructs but GoodApproximations All the Same. J. Electron Spectrosc. Relat. Phenom. 2002,123, 345–363.

(139) Chong, D. P.; Gritsenko, O. V.; Baerends, E. J. Interpretationof the Kohn�ShamOrbital Energies as Approximate Vertical IonizationPotentials. J. Chem. Phys. 2002, 116, 1760–1772.

(140) Luo, J.; Xue, Z. Q.; Liu, W. M.; Wu, J. L.; Yang, Z. Q.Koopmans’ Theorem for Large Molecular Systems within DensityFunctional Theory. J. Phys. Chem. A 2006, 110, 12005–12009.

(141) Zhan, C.-G.; Nichols, J. A.; Dixon, D. A. Ionization Potential,Electron Affinity, Electronegativity, Hardness, and Electron ExcitationEnergy: Molecular Properties from Density Functional Theory OrbitalEnergies. J. Phys. Chem. A 2003, 107, 4184–4195.

(142) Ayers, P.; Melin, J. Computing the Fukui Function from AbInitio Quantum Chemistry: Approaches Based on the Extended Koop-mans’ Theorem. Theor. Chem. Acc. 2007, 117, 371–381.

(143) Shuai, Z.; Wang, L.; Li, Q. Evaluation of Charge Mobility inOrganic Materials: From Localized to Delocalized Descriptions at aFirst-Principles Level. Adv. Mater. 2011, 23, 1145–1153.

(144) Scholes, G. D.; Rumbles, G. Excitons in Nanoscale Systems.Nat. Mater. 2006, 5, 683–96.

(145) Jaiswal, M.; Menon, R. Polymer Electronic Materials: A Re-view of Charge Transport. Polym. Int. 2006, 55, 1371–1384.

(146) K€ohler, A.; dos Santos, D. A.; Beljonne, D.; Shuai, Z.; Br�edas,J.-L.; Holmes, A. B.; Kraus, A.; M€ullen, K.; Friend, R. H. ChargeSeparation in Localized and Delocalized Electronic States in PolymericSemiconductors. Nature 1998, 392, 903–906.

(147) Pohl, H. A. Theories of Electronic Behavior in Macromole-cular Solids. J. Polym. Sci. C: Polym. Symp. 1967, 17, 13–40.

(148) Maitra, N. T. Undoing Static Correlation: Long-RangeCharge Transfer in Time-Dependent Density-Functional Theory.J. Chem. Phys. 2005, 122, 234104.

(149) Peach, M. J. G.; Benfield, P.; Helgaker, T.; Tozer, D. J. Exci-tation Energies in Density Functional Theory: An Evaluation and aDiagnostic Test. J. Chem. Phys. 2008, 128, 044118.

(150) Fabian, J.; Mann, M.; Petiau, M. The Origin of the Color of1,2-Dithiins - Interpretation by Kohn�Sham Orbitals. J. Mol. Model.2000, 6, 177–185.

(151) Hachmann, J.; Dorando, J. J.; Avil�es, M.; Chan, G. K.-L. TheRadical Character of the Acenes: A Density Matrix RenormalizationGroup Study. J. Chem. Phys. 2007, 127, 134309.

(152) Dunitz, J. D.; Gavezzotti, A. HowMolecules Stick Together inOrganic Crystals: Weak Intermolecular Interactions. Chem. Soc. Rev.2009, 38, 2622–2633.

(153) Williams, J. H. The Molecular Electric Quadrupole Momentand Solid-State Architecture. Acc. Chem. Res. 1993, 26, 593–598.

(154) Borsenberger, P. M.; Fitzgerald, J. J. Effects of the DipoleMoment on Charge Transport in Disordered Molecular Solids. J. Phys.Chem. 1993, 97, 4815–4819.

(155) Young, R. H.; Fitzgerald, J. J. Dipole Moments of Hole-Transporting Materials and Their Influence on Hole Mobilityin Molecularly Doped Polymers. J. Phys. Chem. 1995, 99, 4230–4240.

(156) Zade, S. S.; Zamoshchik, N.; Bendikov, M. From ShortConjugated Oligomers to Conjugated Polymers. Lessons from Studieson Long Conjugated Oligomers. Acc. Chem. Res. 2011, 44, 14–24.



(157) Gierschner, J.; Cornil, J.; Egelhaaf, H.-J. Optical Bandgaps ofπ-Conjugated Organic Materials at the Polymer Limit: Experiment andTheory. Adv. Mater. 2007, 19, 173–191.(158) Zade, S. S.; Bendikov, M. From Oligomers to Polymer:

Convergence in the HOMO�LUMO Gaps of Conjugated Oligomers.Org. Lett. 2006, 8, 5243–5246.(159) Ladik, J. J. Recent Developments in the Quantum Theory of

Organic Polymers. J. Mol. Struct. THEOCHEM 1991, 230, 127–142.(160) Ameri, T.;Dennler,G.; Lungenschmied,C.; Brabec,C. J.Organic

Tandem Solar Cells: A Review. Energy Environ. Sci. 2009, 2, 347–363.(161) Heimel, G.; Salzmann, I.; Duhm, S.; Koch, N. Design of

Organic Semiconductors from Molecular Electrostatics. Chem. Mater.2011, 23, 359–377.(162) Troisi, A. Charge Transport in High Mobility Molecular

Semiconductors: Classical Models and New Theories. Chem. Soc. Rev.2011, 40, 2347–2358.(163) Br�edas, J.-L.; Beljonne, D.; Coropceanu, V.; Cornil, J. Charge-

Transfer and Energy-Transfer Processes in π-Conjugated Oligomersand Polymers: A Molecular Picture. Chem. Rev. 2004, 104, 4971–5004.(164) Bethe, H. A.; Salpeter, E. E. Quantum Mechanics of One and

Two Electron Atoms; Plenum: New York, 1977.(165) Singh, R.; Mishra, M. Electronic Structure Analysis and

Vertical Ionization Energies of Thiophene and Ethynylthiophenes.J. Chem. Sci. 2009, 121, 867–872.(166) Smith, M. B.; Michl, J. Singlet Fission. Chem. Rev. 2010, 110,

6891–6936.(167) Houili, H.; Picon, J. D.; Zuppiroli, L.; Bussac, M. N. Polariza-

tion Effects in the Channel of an Organic Field-Effect Transistor. J. Appl.Phys. 2006, 100, 023702.(168) She, C.; Easwaramoorthi, S.; Kim, P.; Hiroto, S.; Hisaki, I.;

Shinokubo, H.; Osuka, A.; Kim, D.; Hupp, J. T. Excess PolarizabilityReveals Exciton Localization/Delocalization Controlled by LinkingPositions on Porphyrin Rings in Butadiyne-Bridged Porphyrin Dimers.J. Phys. Chem. A 2010, 114, 3384–3390.(169) Gilbert, A. T. B.; Besley, N. A.; Gill, P. M. W. Self-Consistent

Field Calculations of Excited States Using the Maximum OverlapMethod (MOM). J. Phys. Chem. A 2008, 112, 13164–13171.(170) Jacquemin, D.; Perp�ete, E. A.; Scalmani, G.; Frisch, M. J.;

Kobayashi, R.; Adamo, C. Assessment of the Efficiency of Long-RangeCorrected Functionals for Some Properties of Large Compounds.J. Chem. Phys. 2007, 126, 144105.(171) Le Bahers, T.; Pauporte, T.; Scalmani, G.; Adamo, C.; Ciofini,

I. A TD-DFT Investigation of Ground and Excited State Properties inIndoline Dyes Used for Dye-Sensitized Solar Cells. Phys. Chem. Chem.Phys. 2009, 11, 11276–11284.(172) Head-Gordon, M.; Grana, A. M.; Maurice, D.; White, C. A.

Analysis of Electronic Transitions as the Difference of Electron Attach-ment and Detachment Densities. J. Phys. Chem. 1995, 99, 14261–14270.(173) Martin, R. L. Natural Transition Orbitals. J. Chem. Phys. 2003,

118, 4775–4777.(174) Hutchison, G. R.; Ratner, M. A.; Marks, T. J. Accurate

Prediction of Band Gaps in Neutral Heterocyclic Conjugated Polymers.J. Phys. Chem. A 2002, 106, 10596–10605.(175) S�anchez-Carrera, R. S.; Atahan, S.; Schrier, J.; Aspuru-Guzik,

A. Theoretical Characterization of the Air-Stable, High-MobilityDinaphtho[2,3-b:2030-f]thieno[3,2-b]-thiophene Organic Semiconduc-tor. J. Phys. Chem. C 2010, 114, 2334–2340.(176) Curtiss, L. A.; Raghavachari, K.; Redfern, P. C.; Pople, J. A.

Assessment of Gaussian-2 and Density Functional Theories for theComputation of Enthalpies of Formation. J. Chem. Phys. 1997, 106,1063–1079.(177) NIST Computational Chemistry Comparison and Bench-

mark Database, NIST Standard Reference Database No 101, Rel. 15a,April 2010; Russell D. Johnson III, Editor; http://cccbdb.nist.gov/(accessed August 10, 2011).(178) Cheung, D. L.; Troisi, A. Modelling Charge Transport in

Organic Semiconductors: From Quantum Dynamics to Soft Matter.Phys. Chem. Chem. Phys. 2008, 10, 5941–5952.

(179) San-Fabi�an, E.; Guijarro, A.; Verg�es, J.; Chiappe, G.; Louis, E.PPP Hamiltonian for Polar Polycyclic Aromatic Hydrocarbons. Eur.Phys. J. B 2011, 81, 253–262.

(180) Verg�es, J. A.; San-Fabi�an, E.; Chiappe, G.; Louis, E. Fit ofPariser�Parr�Pople and Hubbard model Hamiltonians to Charge andSpin States of Polycyclic Aromatic Hydrocarbons. Phys. Rev. B 2010,81, 085120.

(181) Trevethan, T.; Shluger, A.; Kantorovich, L. Modelling Com-ponents of Future Molecular Devices. J. Phys.: Condens. Matter 2010, 22,084024.

(182) World Community Grid Home Page. www.worldcommuni-tygrid.org (accessed August 10, 2011)

(183) Clery, D. IBM Offers Free Number Crunching for Humani-tarian Research Projects. Science 2005, 308, 773.

(184) Shao, Y.; Fusti Molnar, L.; Jung, Y.; Kussmann, J.; Ochsenfeld,C.; Brown, S. T.; Gilbert, A. T. B.; Slipchenko, L. V.; Levchenko, S. V.;O'Neill, D. P.; et al. Advances in Methods and Algorithms in a ModernQuantum Chemistry Program Package. Phys. Chem. Chem. Phys. 2006,8, 3172–3191.

(185) Anderson, D. P. BOINC: A System for Public-Resource Comput-ing and Storage, 2004; http://dx.doi.org/10.1109/GRID.2004.14.

(186) National Energy Research Scientific Computing CenterHome Page. www.nersc.gov (accessed August 10, 2011).

(187) Catlett, C.; Allcock, W. E.; Andrews, P.; Aydt, R.; Bair, R.;Balac, N.; Banister, B.; Barker, T.; Bartelt, M.; Beckman, P.; et al.TeraGrid: Analysis of Organization, System Architecture, and Middle-ware Enabling New Types of Applications. Adv. Parallel Comput. 2008,16, 225–249.

(188) Beazley, D. M. Python Essential Reference, 4th ed.; Addison-Wesley: Upper Saddle River, NJ, 2009.

(189) Langtangen, H. P. Python Scripting for Computational Science,3rd ed.; Springer: Berlin, 2010.

Documents

The Harvard Clean Energy Project: Large-Scale ... · The Harvard Clean Energy Project: ... screening eﬀort toward an active engineering of novel organic ... ideas from machine learning,93,94